Amazon’s AI Spending Is Starting to Look Like Profitable Scale

The Quarter That Changed the Argument

Amazon closed at $271.58 on Friday, July 31, up 15.32% in the first full session after its second-quarter release and earnings call. It finished near the session high on roughly 128 million shares. This was not mere AI enthusiasm. Investors received the clearest operating evidence yet that Amazon’s immense AI infrastructure program is producing scalable commercial returns.

AWS revenue reached $42.232 billion, up 36.7% year over year and roughly 4.3% above Visible Alpha’s pre-earnings consensus. Growth accelerated from 17% to 20%, 24%, 28%, and nearly 37% over five quarters. AWS operating income rose 63.6% to $16.621 billion, while its operating margin expanded to 39.4%. Consolidated operating income reached $27.461 billion, 43.2% above the prior year and $3.461 billion above the top of Amazon’s guidance.

Growth and margin expansion occurring together change the argument. If Amazon were buying expensive computing equipment merely to attract low-quality demand, AWS margins should be deteriorating. Instead, AWS produced about 60.5% of Amazon’s operating income from 21.1% of company sales. The market increased its estimate of how profitably Amazon can turn new capacity into customer workloads.

How Infrastructure Becomes Revenue

That is what the standard bubble critique misses. Amazon must secure land, electricity, buildings, networking equipment, servers, and chips six to 24 months before much of the resulting capacity can be billed. Engineers then convert those inputs into reliable services that customers rent rather than build themselves. As utilization rises, fixed costs are spread across more revenue. Custom silicon can lower unit costs, and the resulting profit helps finance another round of expansion.

Andy Jassy put the relationship plainly in his 2025 shareholder letter: “The faster AWS grows, the more short-term capex we’ll spend.” Management’s claim is not proof, but the quarter supplied evidence for it. Amazon said its AWS AI business exceeded a $25 billion annual revenue run rate and was growing at a triple-digit rate. Bedrock customer spending during the quarter exceeded spending in all previous quarters combined. Its chips business also surpassed a company-defined $25 billion run rate, while Graviton adoption and Trainium commitments suggested that proprietary processors are becoming both customer products and cost-control tools.

The pessimistic frame treats high spending as evidence of managerial irrationality. Here, higher spending is more plausibly the result of stronger demand. Amazon says substantial future capacity already carries customer commitments and expects demand to exceed supply through 2026 and probably 2027. Capacity constraints may delay revenue, but they are not evidence that customers do not want the product. They show that construction is struggling to keep pace with adoption.

The Risks Are Still Real

Trailing-12-month free cash flow fell to negative $7.604 billion as net property-and-equipment purchases reached $169.007 billion. Expected 2026 cash capital expenditure rose from about $200 billion to about $220 billion, partly because of higher memory costs. Third-quarter revenue guidance missed the prevailing consensus. Microsoft, Google, and specialized providers remain formidable competitors. Depreciation, energy costs, pricing pressure, customer concentration, or another investment surge could weaken eventual returns.

Nor does Amazon’s $62.647 billion net-income headline provide clean proof. It included $53.4 billion of non-operating pre-tax income, primarily related to the Anthropic investment. Retail, advertising, broader cloud optimism, and Microsoft’s strong results also helped shape the market response. A 15% rally proves neither perfect causation nor a correct valuation.

What the Quarter Actually Proves

But none of those qualifications supports the claim that AI infrastructure spending is inherently wasteful. AWS growth accelerated, margins expanded, operating income exceeded guidance, and measurable AI usage increased. Management committed capital under uncertainty, engineers created usable capacity, customers bought it, and investors revised their expectations. Future quarters must demonstrate durable utilization and free-cash-flow conversion. This quarter showed, for the first time with real operating force, that Amazon’s AI investment cycle may be moving from costly construction to profitable scale.

MacroShed Markets is informational analysis, not investment advice.